M2LSH:基于LSH的高维数据近似最近邻查找算法

在许多应用中,LSH(Locality Sensitive Hashing)以及各种变体,是解决近似最近邻问题的有效算法之一.虽然这些算法能够很好地处理分布比较均匀的高维数据,但从设计方案来看,都没有针对数据分布不均匀的情况做相应的优化.针对这一问题,本文提出了一种新的基于LSH的解决方案(M2LSH,2 Layers Merging LSH),对于数据分布不均匀的情况依然能得到一个比较好的查询效果.首先,将数据存放到具有计数功能的组合哈希向量表示的哈希桶中,然后通过二次哈希将这些桶号投影到一维空间,在此空间根据各个桶中存放的数据个数合并相邻哈希桶,使得新哈希桶中的数据量能够大致均衡.查询时仅访问有限个哈希桶,就能找到较优结果.本文给出了详细的理论分析,并通过实验验证了M2LSH的性能,不仅能减少访问时间,也可提高结果的正确率.     

一种安全鲁棒的图像哈希方法*

互联网的发展使得多媒体的版权保护成为亟待解决的问题,从而使得用于图像内容认证的哈希方法得到广泛的研究和应用。提出了一种新的基于V系统的图像哈希方法,该方法首先将原始图像强化,然后提取V变换域上的系数生成哈希值。实验结果显示,该方法能够抵抗滤波、噪声、25%以下的裁剪和95%以下的JPEG压缩等攻击,有很好的安全性和鲁棒性。     

基于视觉的工业机器人异常动作检测方法研究

工业机器人的突发故障引发的安全问题时有发生。传统的基于数据分析的故障诊断方法存在传感器数据易受干扰,机器人通讯协议不统一,监测系统嵌入在执行系统内部相互影响等问题。提出一种基于机器视觉的工业机器人故障动作检测方法。对工业机器人作业视频进行实时分析,采用图像分割技术分离工业机器人本体并采用图像哈希技术生成工业机器人姿态编码,结合序列模式分析技术检测工业机器人异常动作并进行预警。不依赖于工业机器人通讯协议,以非接触式的方式对工业机器人进行实时监控,具有易于部署和成本低的特点。基于自主构建的工业机器人仿真视频数据集进行了实验研究,结果表明提出的方法可准确识别工业机器人异常动作,精确率和召回率均为100%。     

基于散列布尔矩阵的关联规则Eclat改进算法*

将散列表与布尔矩阵相结合,提出了一种基于散列布尔矩阵的Eclat改进算法,通过提高求交集的速度来加快整个算法生成频集的过程。实验结果表明,改进的Eclat算法在计算性能和时间效率上均优于传统算法。     

Multidimensional extendible hashing for partial-match queries

Hashing is a well-known technique for organizing direct access files. Extendible hashing removes the restriction on the expansion of the file and thus allows dynamic files. We generalize the technique to store multi-attribute keys. Exact-match queries (searching) can be done in constant time usingn-dimensional hashing. Ann-dimensional partial-match queries givenk attributes can be answered inO(N**((n –k)/n)) time whereN is the number of records stored. It is shown thatn-dimensional hashing is a special case of one-dimensional hashing, thus the storage utilization of the buckets is independent ofn. Simulation results are presented to show the advantages of multidimensional hashing.This research was partially supported by a Research Initiation Grant from the University of Houston.     

A Review of Audio Fingerprinting

An audio fingerprint is a compact content-based signature that summarizes an audio recording. Audio Fingerprinting technologies have attracted attention since they allow the identification of audio independently of its format and without the need of meta-data or watermark embedding. Other uses of fingerprinting include: integrity verification, watermark support and content-based audio retrieval. The different approaches to fingerprinting have been described with different rationales and terminology: Pattern matching, Multimedia (Music) Information Retrieval or Cryptography (Robust Hashing). In this paper, we review different techniques describing its functional blocks as parts of a common, unified framework. Pedro Cano received a B.Sc and M. Sc. Degree in Electrical Engineering from the Universitat Politècnica de Catalunya in 1999. In 1997, he joined the Music Technology Group of the Universitat Pompeu Fabra where he is currently pursuing his Ph.D. on Content-based Audio Identification. He has been assistant professor in the Department of Technologies of the Universitat Pompeu Fabra since 1999. His research interests and recent work include: signal processing for music applications, within a real-time voice morphing system for karaoke applications, pattern matching and information retrieval, specifically content-based audio identification. Eloi Batlle received his M.S. degree in electronic engineering in 1995 from the Politechnical University of Catalunya in Barcelona, Spain. He then joined the Signal Processing Group at the same university where he was working on robust speech recognition. He received a PhD on this subject in 1999. While he was a PhD student he also worked as a researcher at the Telecom Italia Lab during 1997. In 2000 he joined the Audiovisual Institute (a part of the Pompeu Fabra University). Currently he is a member of the Music Technology Group of the same Institute where he leads several reseach projects on music identification and similarity. In 2000 he also joined the Department of Technologies of the Pompeu Fabra University and he teaches several subjects to undergraduate and graduate students. From 2001 he is the Deputy Director of this Department. His research interests include information theory, music similary, statistical signal processing and pattern recognition. Ton Kalker was born in The Netherlands in 1956. He received his M.S. degree in mathematics in 1979 from the University of Leiden, The Netherlands. From 1979 until 1983, while he was a Ph.D. candidate, he worked as a Research Assistant at the University of Leiden. From 1983 until December 1985 he worked as a lecturer at the Computer Science Department of the Technical University of Delft. In January 1986 he received his Ph.D. degree in Mathematics. In December 1985 he joined the Philips Research Laboratories Eindhoven. Until January 1990 he worked in the field of Computer Aided Design. He specialized in (semi) automatic tools for system verification. Currently he is a member of the Processing and Architectures for Content MANagement group (PACMAN) of Philips Research, where he is working on security of multimedia content, with an emphasis on watermarking and fingerprinting for video and audio. In November 1999 he became a part-time professor in the Signal Processing Systems group of Jan Bergmans in the area of ‘signal processing methods for data protection’. He is a Fellow of the IEEE for his contributions to practical applications of watermarking, in particular watermarking for DVD-Video copy protection. His other research interests include wavelets, multirate signal processing, motion estimation, psycho physics, digital video compression and medical image processing. Jaap Haitsma was born in 1974 in Easterein, the Netherlands. He received his B.Sc. in Electronic Engineering from the Noordelijke Hogeschool Leeuwarden in 1997. He did his thesis in 1997 at the Philips Research Laboratories in Redhill, England, on the topic of: “Colour Management for Liquid Crystal Displays”. Currently he is with the Philips Research Laboratories, Eindhoven, the Netherlands, where he has been doing research into digital watermarking and fingerprinting of audio and video since late 1997. From 1999 to 2002 he was also a part-time student at the Technical University of Eindhoven, where he obtained his M.Sc. in Electronic Engineering. His areas of interest include digital signal processing, database search algorithms and software engineering.